AI Code Governance: Frequently Asked Questions
This FAQ specifically covers Sigrid’s AI Generated Code Detection functionality. Also check the Sigrid FAQ for more information about Sigrid in general.
How does Sigrid detect AI generated code?
Sigrid uses a machine learning classification model that is trained on a set of stylometric features of AI generated and human-written code. Examples of stylometric features are line length, number of blank lines, indentation, casing, and similar source code characteristics.
Sigrid takes into account over 90 different stylometric features and is able to identify AI generated code with an accuracy of over 95%.
Does Sigrid use AI models on my code to perform detection?
Sigrid uses a self-trained local machine learning classification model on code metrics derived from your source code. Sigrid does not send your code to a public AI service such as ChatGPT.
Which programming languages are supported?
See the Technology Support section.
Which coding assistants are supported?
Sigrid’s AI Generated Code Detection algorithm is trained on code generated by various AI coding assistants, such as:
- Claude Code
- Cursor
- GitHub Copilot
- OpenAI Codex
- JetBrains Junie
- Windsurf
- Kiro
- AntiGravity
- Amp
At which granularity level does Sigrid identify AI generated code?
Sigrid identifies AI generated code at the level of units, which are methods and functions.
Methods and functions that are smaller than 5 lines of code are ignored, because they do not provide enough signal to determine whether they were AI generated or human written.
Why does Sigrid identify fragments of code that have been manually written as AI generated?
Sigrid uses a machine learning classification model with an accuracy of over 95%. This means that roughly 19 out of 20 code fragments are classified correctly. In large codebases, this can still result in a noticeable number of false positives being identified as AI generated.
The goal of AI Generated Code Detection is not to achieve perfect accuracy, but rather to provide useful insights into current adoption of AI coding and related trends. For that purpose, the current detection capability is sufficiently accurate.
Can I report wrongly detected code somewhere?
Yes, you are welcome to submit a link to an AI generated code finding to our support team. We will use your feedback to improve the next generation of our detection model.
Please be aware that the aim of the detection approach is not to be 100% accurate, but rather to be accurate enough to provide enterprises with meaningful insights into AI coding adoption trends in their software portfolio.
Contact and support
Feel free to contact SIG’s support team for any questions or issues you may have after reading this documentation or when using Sigrid.